Human Stress Detection in and through Sleep – A Deep Learning Approach

Stress refers to the man's ability to respond to any external or internal threat or pressure and has a direct or indirect impact on one's health. Many health-related disorders are linked to stress and thus monitoring, measuring, and managing stress is simply a lifesaving remedy. Due to the automatic ability of deep neural networks to learn features, this study is proposed based on the four-layer Feed Forward Neural Network (4-FFNN) model for the detection and classification of physiological signals obtained from a smart yoga pillow (SaYoPillow) into five distinct classes: namely: normal, medium low, medium, medium high, high. The proposed FFNN model consists of an input layer to accept 8 distinct features followed by 2 hidden layers back-to-back and lastly an output SoftMax layer of 5 neurons. The accuracy of the proposed stress detection model is compared with a few machine learning models as well as other stress detection-based studies. The average accuracy of 99.9% is certainly better than the traditional approaches for detecting stress and the usage of the smart pillow makes the model more robust and easier to use.